RSS/Atom feed Twitter
Site is read-only, email is disabled
  1. Chat
  2. Useful links
  3. Polls

Featured content

Ubuntu 10.10: GPU hardware acceleration (OpenCL) available

By at 2010-10-03 20:54:40 UTC, last updated over 13 years ago. CC BY-NC License

Ubuntu 10.10 comes with OpenCL support. OpenCL is a language that allows to access the GPU and other parallel processors. That means that the graphics card’s shaders can be used from within any application. Multimedia or image processing applications like GIMP or GIMP plugins are perfect candidates for this optimisation that is often just called “hardware acceleration”.

If you decide to install the nvidia proprietary graphics card drivers (you can do this with a few clicks), CUDA (the Nvidia hardware acceleration API) and OpenCL support is installed too. [Information for ATI/fglx still needed, can you help? Please comment.]

There’s also a package for PyOpenCL called python-pyopencl which allows developers to get full access to OpenCL using Python.

That means that future GIMP plugins could be written in Python with OpenCL and hardware acceleration support, bringing incredible performance gains (1000 times faster and even more). I am glad and hopeful to see first applications or plugins that use this great new feature soon.

Example output from the PyOpenCL benchmark script:

python /usr/share/doc/python-pyopencl/examples/benchmark-all.py
Execution time of test without OpenCL:  14.9085888863 s
===============================================================
Platform name: NVIDIA CUDA
Platform profile: FULL_PROFILE
Platform vendor: NVIDIA Corporation
Platform version: OpenCL 1.0 CUDA 3.1.1
---------------------------------------------------------------
Device name: GeForce 9300M GS
Device type: GPU
Device memory:  255 MB
Device max clock speed: 1450 MHz
Device compute units: 1
Execution time of test: 0.0199275 s
Results OK

Comments

Post your own comments, questions or hints here. The author and other users will see your posting and can reply to it.

Of course, you can also ask in the chat.

Subscription management

Please log in to manage your subscriptions.

User rating

This topic (Ubuntu 10.10: GPU hardware acceleration (OpenCL) available) has been rated 2.0/5.0.

New comments are disabled because of spam.

gorbalad
over 13 years ago

Execution time of test without OpenCL: 6.56879401207 s
===============================================================
Platform name: NVIDIA CUDA
Platform profile: FULL_PROFILE
Platform vendor: NVIDIA Corporation
Platform version: OpenCL 1.0 CUDA 3.2.1
---------------------------------------------------------------
Device name: GeForce GTX 460
Device type: GPU
Device memory: 1023 MB
Device max clock speed: 1648 MHz
Device compute units: 7
Execution time of test: 0.000532896 s
Results OK

need I say more? :)

redforce member for over 17 years redforce 230 comments
over 13 years ago

To activate: install nvidia drivers via Ubuntu

To check: install python-pyopencl and do:
python /usr/share/doc/python-pyopencl/examples/benchmark-all.py

geoffrey member for over 13 years geoffrey 1 comment rated this topic with 2/5
over 13 years ago

How is OpenCL activated or launched. No evidence
in my copy of 10.10 Ubuntu

Geoffrey

lamapper
over 13 years ago

Makes me glad I bought a Linux PC from ZaReason with 4 cores, Nvidia GPU built in, and with 4GB of RAM on Linux...its already faster then most other alternatives.

And I just updated to Ubuntu 10.10 this week, yea!

Imagine the speed increases with multiple cores, over multiple hours with the right coding algorithms....very nice.

Imagine this capability in a Linux server farm for working with video. This is very nice news indeed.

Vincent of StreamComputing.eu
over 13 years ago

A plugin consists of a configuration/GUI/wrapper/etc and a computation; only the computation needs to be OpenCL'ed. On Windows CPU-emulation is possible, but it's (still) slower than native. So the computation needs to be written in OpenCL, in addition to the CPU-code. It is extra work, but many OpenCL-kernels for graphics (like FFT) have already been developed.

Assume we "only" get a speedup of 10x; that's 6 seconds instead of one minute.

NotZed
over 13 years ago

For ATI you just need to install the AMD Stream SDK. Since it also supports CPU for OpenCL it isn't included in the AMD drivers on it's own (so far?).

1000x wouldn't be typical for otherwise optimised code (plain python is another matter), but in the 100s is possible with many graphics algorithms.

devvv member for over 17 years devvv 1475 comments
over 13 years ago

I agree, nevertheless - the possibilities to make GIMP much more performant seems to be easier now which is really good news. However I don't know if there is a "fallback" currently to cpu rendering if there are no gpu shaders present.

Vincent of StreamComputing.eu
over 13 years ago

Nice, but 1000 times seems a bit exaggerated; the results of this specific test is no guarantee for more complex operations in The Gimp.
For Ubuntu 10.04 users, install the following packages (Netherlands-based mirrors, but files are small):
http://nl.archive.ubuntu.com/ubuntu/pool/main/b/boost1.42/libboost-python1.42.0_1.42.0-3ubuntu1_i386.deb
http://nl.archive.ubuntu.com/ubuntu/pool/universe/p/pytools/python-pytools_10-7_all.deb
http://nl.archive.ubuntu.com/ubuntu/pool/multiverse/p/pyopencl/python-pyopencl_0.92~beta+git20100709-1ubuntu1_i386.deb